本论文的主要目的是研究我国债券二级市场中信用债信用利差的主要影响因素，分析哪些特征指标对信用利差有较强的相关性和指示意义。因为债券二级市场常把中债估值当作债券收益率的公允值，因此本文使用信用债中债估值减去同剩余期限的国债收益率作为信用利差因变量。为获得同剩余期限国债的收益率，本文采用Nelson-Siegel模型对国债收益率期限结构曲线进行拟合。本文信用利差研究选取的信用债标的主要为二级市场中存续的中期票据。因为我国城投债和产业债的背景差别较大，因此将中期票据区分为城投债和产业债分别进行研究。然后从债项、评级、宏观、偿债能力、资本结构、盈利能力、营运能力、成长能力和财务评分等多方面选取可能与信用利差相关的七十多个指标作为实证模型的解释变量。本文选取XGBoost模型作为解释变量和信用利差之间的拟合模型。XGBoost模型原理较为成熟，且在学术界和产业界得到广泛应用。SHAP是目前比较先进且被广泛使用的机器学习可解释性算法。本文在用XGBoost建立信用利差模型后，用SHAP分析各解释变量的重要程度及和信用利差的相关模式。本文实证过程中，将数据集按4：1的比例划分训练集和测试集，用训练集数据训练模型，用测试集数据检验误差收敛情况。然后，基于SHAP可解释性算法对模型结果进行分析。根据模型结果，主体评级YY、省份、票面利率、剩余期限、国债10年_100周期均值偏离、R007的20日和50日均值、流动负债合计、现金比率等是影响城投债券信用利差的前十大因素。主体评级YY、票面利率、发行总额、公司属性、省份、国债10年_100周期均值偏离、所属Wind行业名称、剩余期限、R007的20日均值、流动负债合计等是影响产业债信用利差的前十大因素。对于未来研究的展望，主要是用相关方法指导信用债投资。一种是借鉴股票多因子量化投资模型的思路。另一种是价值投资思路，判断模型预测值低于实际信用利差的信用债是否有投资价值。

The main purpose of this thesis is to study the main influencing factors of credit spreads in the Chinese bond market, and analyze which indicators have strong relevance and indicating significance to credit spreads. Because the bond market often regards the valuation of ChinaBond as the fair value of bond yield, this paper uses the valuation of ChinaBond minus the yield of Treasury bonds with the same residual maturity as credit spread. In order to obtain the yield of Treasury bonds with the same residual maturity, this paper adopts Nelson-Siegel model to fit the term structure curve of Treasury bond .In this paper, medium-term notes are selected to study. Because the background is quite different, the medium-term notes are divided into Chinese municipal bonds and industrial bonds respectively. Then, from the aspects of debt, rating, macro, solvency, capital structure, profitability, operating capacity, growth capacity and financial score, more than 70 indicators are selected as explanatory variables of the empirical model. This paper chooses XGBoost model as the fitting model between explanatory variables and credit spreads. XGBoost model is mature, and has been widely used in academia and industry. SHAP is an advanced and widely used machine learning interpretability algorithm. After establishing the credit spread model with XGBoost, SHAP is used to analyze the importance of each explanatory variable and the correlation with credit spread. In this paper, the data set is divided into the training set and the test set according to the ratio of 4:1, the training set data is used to train the model, and the test set data is used to check the error convergence. Then, the model results are analyzed based on SHAP interpretability algorithm. According to the model results, the rating of YY, province, coupon rate, residual maturity, 100 cycle mean deviation of 10-year Treasury bond, 20-day and 50-day mean of R007, total current liabilities, cash ratio, etc. are the top ten factors affecting the credit spread of Chinese municipal bonds. the rating of YY, coupon rate, total issuance, company attributes, province, 100 cycle mean deviation of 10-year Treasury bond, industry name in Wind, residual maturity, 20-day mean of R007, total current liabilities, etc., are the top ten factors affecting the credit spread of industrial bonds. As for the prospect of future research, it is mainly to guide the bond investment by relevant methods. One is to learn from the stock multi-factor quantitative investment model. The other is value investment method, judging whether the credit bond whose predicted value of the model is lower than the actual credit spread has investment value.